Study On The Breakout Prediction Model Based On GA-BP Neural Network

Breakout is the most harmful production accident in the continuous casting process. Once the breakout happens during the performance, the whole process will be forced to stop resulting in hindrance of the normal steelmaking process. It is also detrimental to the stability of production, quality of products, safeties of workers and service life of equipments. One of the most common patterns of breakouts is sticking type between initial solidification shell and mold wall. There are now two theoretical and practical ways to avoid or decrease the breakout accident. On the one hand, researchers are trying to find out the mechanism of breakout and then forbid the responsible technological parameters during the process. On the other hand, people are working hard to develop the prediction system for breakout, measuring the breakout symptom of the mold during the continuous cast process and correspondingly take the measurements of deceleration to avoid the breakout accident. Accordingly, the formation mechanism of sticking-type breakout, measuring methods and prediction principles were investigated in the dissertation. Based on the former studies, the method of breakout prediction was also studied with the introduction of neural network technology through thermocouple thermometry in order to search for an efficient and feasible model of breakout prediction.On the support of the technology creation project named Integration and creation for high quality slab continuous casting process, a system of breakout prediction based on the GA-BP neural network is developed by the investigation of BP (Back Propagation) neural network as well as GA (Genetic Algorithm) and the analysis of characteristic change of thermocouples temperature during sticking-type breakout to eventually construct a model of breakout prediction for slab continuous casting process. The model is composed of a single couple sequence network and a group couple space network. The single couple sequence network is used to identify the temperature change of one same thermocouple with time while the group couple space network is used to distinguish the characteristic heat transfer of different thermocouples towards different directions. Then, a breakout signal is alarmed after the judgment on the output from the space network.The visual design of the breakout prediction model is implemented with program languages of Visual Basic and Matlab, and combined with the history date recorded from the continuous casting field of a steel plant, both the conventional model of BP neural network and the developed model of GA-BP neural network were applied to the breakout prediction process. Testing samples are58groups in amount which includes34groups of breakout alarming samples,26groups of normal casting samples and8groups of temperature change under unsteady state. The simulation results show that the developed model of GA-BP neural network is able to distinguish the typical temperature change patterns of sticking-type breakout process more precisely with97.4%forecast rate and100%alarming rate. The accuracy of the developed model can effectively and efficiently predict the breakout accident. It is found, through the analysis of simulation results, that the false alarm always happens during the unsteady process of the beginning of continuous casting and swift change of tundish because the temperature change patterns of the mold are similar to the sticking-type breakout which eventually results in disturbance to prediction model.